The impact of fiscal policy on

economic activity over the business cycle

evidence from a threshold VAR analysis
Anja Baum
(University of Cambridge)
Gerrit B. Koester
(Deutsche Bundesbank)
Discussion Paper
Series 1: Economic Studies
No 03/2011
Discussion Papers represent the authors personal opinions and do not necessarily reflect the views of the
Deutsche Bundesbank or its staff.
Editorial Board: Klaus Dllmann
Frank Heid
Heinz Herrmann
Karl-Heinz Tdter
Deutsche Bundesbank, Wilhelm-Epstein-Strae 14, 60431 Frankfurt am Main,
Postfach 10 06 02, 60006 Frankfurt am Main
Tel +49 69 9566-0
Telex within Germany 41227, telex from abroad 414431
Please address all orders in writing to: Deutsche Bundesbank,
Press and Public Relations Division, at the above address or via fax +49 69 9566-3077
Internet http://www.bundesbank.de
Reproduction permitted only if source is stated.
ISBN 978-3865586902 (Printversion)
ISBN 978-3865586919 (Internetversion)
Abstract:
Does the state of the business cycle matter for the eects of scal policy
shocks on GDP? This study analyses quarterly German data from 1976 to
2009 in a threshold SVAR, expanding the SVAR approach by Blanchard and
Perotti (2002). In a linear benchmark SVAR, the analysis nds that hiking
spending yields a short-term scal multiplier of around 0.70, while the scal
multiplier resulting from an increase in taxes and social security contributions
is -0.66. In addition, the threshold model derives fundamentally new insights
on the eects of shocks, depending on when in the business cycle they occur,
their size and their direction. Most importantly, scal spending multipliers
are much larger in times of a negative output gap but have only a very limited
eect in times of a positive output gap. Discretionary revenue policies, on
the other hand, have a generally more limited impact. Our ndings have
important implications for the optimal scal policy mix over dierent stages of
the business cycle. Various robustness checks, including a dierent threshold
specication, do not inuence these implications substantially.
Keywords: scal policy, business cycle, nonlinear analysis, scal multipliers
JEL-Classication: E62, E32, C54.
Non-technical summary
What are the eects of scal policy on economic growth? How eective
is it in smoothing the business cycle? Does the state of the business cycle
matter for the eects of scal policy shocks on GDP? These policy-relevant
macroeconomic questions are highly controversial, and the optimal scal policy
action with respect to the size, timing and the policy mix is the topic of erce
debate in the literature.
This paper seeks to contribute to the empirical literature on scal policy
in Germany by adding an additional dimension to the usual linear analysis:
allowing for asymmetries of scal policy shocks on growth depending on their
size, their direction and their timing with respect to the business cycle. We
apply a threshold VAR approach, which is characterised by a separation of the
observations into dierent regimes based on a threshold variable, to model time
series non-linearities. Within each regime, the time series is then assumed to be
described by a linear model. In the baseline specication, we use the output
gap as the threshold variable as it divides economic development in phases
of under- and overutilisation the two regimes under which we expect the
eects of scal stimuli to dier. To identify discretionary scal policy shocks,
we employ exogenously determined elasticities for the working of automatic
stabilisers.
Our research shows that short-term scal multipliers in Germany are in
general moderate and that the state of the business cycle strongly matters for
the eects of scal policy shocks. In a linear benchmark model, the analy-
sis nds that the eect of reductions in tax and social security contributions
and of increased spending on GDP each corresponds to a short-term scal
multiplier of around 0.7, putting it in the range of other empirical results for
Germany. In addition the threshold model derives fundamentally new insights
on the eects of shocks, depending on their timing with respect to the business
cycle. Most importantly, scal spending multipliers are much larger in times of
a negative output gap but have only a very limited eect in times of a positive
output gap. Discretionary revenue policies, on the other hand, have generally
a more limited eect. With respect to the cycle, their impact is larger in the
upper than in the lower output gap regime. Various robustness checks, in-
cluding a dierent threshold specication, do not inuence the resulting policy
implications substantially.
Nichttechnische Zusammenfassung
Wie beeinusst die Fiskalpolitik das Wirtschaftswachstum? Wie eektiv
ist ihr Einsatz zur Glattung des Konjunkturzyklus? Unterscheiden sich die
Eekte von skalpolitischen Stimuli abhangig von der aktuellen Auslastung
einer

Okonomie? Diese wirtschaftspolitisch relevanten Fragen werden kontro-
vers diskutiert und uber die optimale Ausgestaltung skalpolitischer Impulse
hinsichtlich ihres Umfangs, ihrer Terminierung und der verwendeten Instru-
mente wird in der Literatur heftig gestritten.
Ziel dieses Papieres ist es, den ublicherweise in der Literatur zu nd-
enden linearen Analysen der deutschen Fiskalpolitik eine weitere Dimension
hinzuzuf ugen, welche asymmetrische Reaktionen auf skalpolitische Schocks
abhangig von ihrer Gr oe, ihrer Richtung und und ihrer Terminierung im
Konjunkturzyklus erlaubt. Dazu sch atzen wir ein vektorautoregressives
Schwellenwert-Modell, welches die Analyse nicht-linearer Eekte durch
eine Einteilung der empirischen Beobachtungen in zwei unterschiedliche, in
Abhangigkeit von einem Schwellenwert denierte Regime erm oglicht. Inner-
halb jedes dieser zwei Regime wird dann ein lineares Modell angenommen. Im
Basismodell verwenden wir die Produktionsl ucke als Schwellenwert, da diese
den Konjunkturzyklus in Phasen der Unter- und der

Uberauslastung aufteilt
- jene beiden Regime, in denen wir unterschiedliche Eekte von Fiskalstimuli
erwarten. Um die diskretion aren skalischen Schocks zu identizieren verwen-
den wir exogen bestimmte Elastizit aten, die das Wirken der automatischen
Stabilisatoren abbilden. Unsere Analyse zeigt, dass die kurzfristigen Fiskal-
multiplikatoren in Deutschland generell begrenzt sind und dass die jeweilige
Position im Konjunkturzyklus einen wichtigen Einuss auf die Wirksamkeit
von Fiskalstimuli hat.
In einem linearen Referenzmodell ergibt sich f ur K urzungen von Steuern
und Sozialabgaben und f ur Ausgabenerh ohungen zun achst ein kurzfristiger
Multiplikator von rund 0,7 - was im Bereich der Ergebnisse ahnlicher Studien
f ur Deutschland liegt. Unser Schwellenwert-Modell ermoglicht dar uber hinaus
grundlegend neue Einsichten in die Eekte von Schocks in Abh angigkeit von
ihrer Terminierung im Konjunkturzyklus. Wichtigstes Ergebnis ist dass die
Ausgabenmutliplikatoren in Zeiten der Unterauslastung deutlich gr oer sind
als in Zeiten der

Uberauslastung. Diskretion are Einnahmenschocks dagegen
haben einen insgesamt geringeren Eekt als Ausgabenschocks. Hinsichtlich
der Terminierung im Konjunkturzyklus ist der Eekt von Einnahmenschocks
im oberen Regime gr oer als im unteren Regime. Diese Ergebnisse erweisen
sich als robust gegen uber zahlreichen getesteten Modikationen des Modells -
einschlielich einer anderen Spezizierung des Schwellenwertes.
Contents
1 Introduction 1
2 Empirical Literature 4
3 Methodology 8
4 The Data 12
5 Estimation 16
5.1 Model Specication . . . . . . . . . . . . . . . . . . . . . . . . . 16
5.2 Threshold Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
5.3 Identication . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
5.4 Impulse Response Analysis . . . . . . . . . . . . . . . . . . . . . 21
5.4.1 Linear Impulse Response . . . . . . . . . . . . . . . . . . 21
5.4.2 Lower Regime, 2% Growth Shock . . . . . . . . . . . . . 22
5.4.3 Upper Regime, 2% Growth Shock . . . . . . . . . . . . . 23
5.4.4 Comparison Lower and Upper Regime, Increasing Shock
Size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
6 Robustness Checks 25
6.1 GDP Growth Threshold . . . . . . . . . . . . . . . . . . . . . . 25
6.2 Structural Identication . . . . . . . . . . . . . . . . . . . . . . 26
6.3 Data Sample and Threshold Value . . . . . . . . . . . . . . . . . 27
7 Conclusions 28
A GIRF Algorithm 30
B Exogenous Elasticities 31
C Literature 33
List of Tables
1 German VAR Analysis, Impact of Fiscal Shocks on GDP . . . . 7
2 Unit Root Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3 Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . 16
4 Tsay Threshold Test . . . . . . . . . . . . . . . . . . . . . . . . 17
5 Fiscal Multipliers . . . . . . . . . . . . . . . . . . . . . . . . . . 25
6 Calculated Elasticities . . . . . . . . . . . . . . . . . . . . . . . 31
7 Elasticities in the literature . . . . . . . . . . . . . . . . . . . . 32
List of Figures
1 Growth Rates of Series . . . . . . . . . . . . . . . . . . . . . . . 15
2 Output Gap Regimes, Threshold Value=0.15% . . . . . . . . . . 18
3 Time-Varying Elasticities . . . . . . . . . . . . . . . . . . . . . . 19
4 Adjusted Revenue - Output Gap Correlation . . . . . . . . . . . 19
5 Linear Impulse Responses, Shocks in R and G . . . . . . . . . . 22
6 Lower Regime: 2% Growth Shock . . . . . . . . . . . . . . . . . 23
7 Upper Regime: 2% Growth Shock . . . . . . . . . . . . . . . . . 24
8 Lower Regime: a
1
= 0.5, 2% Growth Shock . . . . . . . . . . . . 39
9 Lower Regime: a
1
= 1.5, 2% Growth Shock . . . . . . . . . . . . 39
10 Lower Regime: 1976-2008 . . . . . . . . . . . . . . . . . . . . . 40
11 Upper Regime: 1976-2008 . . . . . . . . . . . . . . . . . . . . . 40
12 5% Growth Shock - Lower Regime . . . . . . . . . . . . . . . . . 41
13 5% Growth Shock - Upper Regime . . . . . . . . . . . . . . . . 41
14 IR for GDP Growth as Threshold: 2% Growth Shock - Lower
Regime . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
15 IR for GDP Growth as Threshold: 2% Growth Shock - Upper
Regime . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
16 Lower Regime: Threshold of Zero - 2% Growth Shock . . . . . . 43
17 Upper Regime: Threshold of Zero - 2% Growth Shock . . . . . . 43
18 Lower Regime: Cholesky Order GDP R G . . . . . . . . . 44
19 Lower Regime: Cholesky Order R G GDP . . . . . . . . . 44
20 Upper Regime: Cholesky Order GDP R G . . . . . . . . . 45
21 Upper Regime: Cholesky Order R G GDP . . . . . . . . . 45
The Impact of Fiscal Policy on Economic
Activity over the Business Cycle
-
Evidence from a Threshold VAR Analysis
1
1 Introduction
What are the eects of scal policy on economic growth? How eective
is it in smoothing the business cycle? Does the state of the business cycle
matter for the eects of scal policy shocks on GDP? These policy-relevant
macroeconomic questions are highly controversial, and the optimal scal policy
action with respect to the size, timing and the policy mix is the topic of erce
debate in the literature.
The nancial turmoil in 2008/2009 has further strengthened the interest
of governments, central banks and academia in the role of scal policy. The
traditional monetary transmission mechanism is weak and monetary policy
alone seems unable to counter the huge contraction of demand. Furthermore,
many countries have nearly reached the zero lower bound, with no more room
to reduce central bank interest rates further. As a consequence, huge scal
stimulus packages have been introduced in Germany as well as in most in-
dustrialised countries worldwide. And although the belief is strong that these
packages helped many countries to recover from the crisis, our knowledge of
the eectiveness of scal stimuli in such downturns based on the theoretical
and empirical literature is still very limited. This paper seeks to contribute
to the empirical literature on scal policy in Germany by adding an additional
dimension to the usual linear analyses, allowing for asymmetries of scal policy
shocks on growth depending on their size, their direction and their timing with
respect to the business cycle.
In the theoretical literature we nd strongly diverging positions with re-
spect to the general eectiveness of scal stimuli. For example, standard Real
Business Cycle (RBC) models expect an increase in government consumption
to be completely oset by a reduction of private consumption (see Baxter and
1
Authors: Anja Baum, University of Cambridge, email: ab862@hermes.cam.ac.uk; Ger-
rit B. Koester, Deutsche Bundesbank, email: gerrit.koester@bundesbank.de. The views
expressed in this paper are those of the authors and do not necessarily reect the opinions
of the Deutsche Bundesbank or its sta. We thank Carsten Trenkler, Patra Geraats, Malte
Kn uppel, J org Breitung, Karsten Wendor, Christoph Priesmeier and Heinz Herrmann for
helpful comments and suggestions. The usual disclaimer applies.
1
King 1993, Christiano and Eichenbaum 1992 or Fat as and Mihov 2001). On
the other hand, standard Keynesian models argue that consumers are non-
Ricardian
2
and a government consumption shock increases private consump-
tion and GDP (Blanchard 2001). However, so far little research eort has
been spent on the question of whether the eectiveness of scal policy might
vary depending on macro-economic circumstances. These eects can best be
covered in a non-linear policy analysis.
3
There are several reasons why the reaction to scal stimuli can be non-
linear. Looking at the supply side of the economy, we can distinguish periods of
positive and negative output gaps. The traditional crowding out argument (see
Buiter 1977), stating that government expenditure replaces private spending,
is generally applicable in times of a positive output gap but less so in times
when output is below potential and excessive capacities in the economy are
available. This gives scal policy the chance to activate unused factors of
production.
4
We also nd several arguments for a non-linear impact analysis on the
demand side. For example, Drazen (1990) argues that the eects of scal
policy depend on the size and persistence of the scal impulse, because both
inuence the signalling eect with respect to the scal policy that is to be
expected in the future (see Giavazzi et al. (2000) for empirical support).
5
Additionally, in times of high negative output gaps and high unemployment
individuals and rms are facing tighter credit constraints, as banks eliminate
credit lines or increase the risk premia on interest rates for loans. Severely
credit-constrained borrowers tend to adjust spending substantially in response
to even a contemporaneous change in disposable income, which can even result
2
The importance of non-Ricardian households for scal policy eects is discussed in
Coenen and Straub (2005), among others.
3
Corsetti et al. (2010) argue in favour of non-linear eects of scal policy, with especially
high scal multipliers after strong recessions. Christiano et al. (2009) state that scal policy
is most eective in the case of very low interest rates (which are likely to occur in times of
high negative output gaps).
4
Along these lines, Kn uppel (2008) analyses the consequences of an inclusion of capacity
constraints in the RBC framework by means of a Markov switching model. He argues that
those capacity constraints are not binding in recessions, leaving economic agents more room
to react to policy measures.
5
The strong eects of the German governments massive scal stimulus packages
including the cash for clunkers (Abwrackpr amie) program as well as nancial market
stabilisation and the guarantee of deposits during the 2008/09 economic downswing can be
see as examples for such a signalling.
2
from a change in interest rates (see for example J a askela 2007). Although
the analysis of credit constraints has been applied mostly to monetary policy
research (see for example Blinder 1987, Galbraith 1996, Weise 1999, Balke
2000 and Calza and Sousa 2006), there is little reason to doubt that scal
policy can inuence disposable income and thus consumption especially that
of credit-constrained households by tax cuts or by increases in transfers to
the most severely credit-constrained consumers and rms (for a discussion see
Gal` et al. 2007 or Roeger and int Veld 2009).
If any of the arguments for non-linearity applies, the linear VAR framework,
which dominates the empirical literature, is not adequate. Tying a non-linear
economy to a linear VAR framework can lead to misleading inferences with
respect to its dynamics. Several approaches to model time series non-linearities
can be found in the literature, including Markov switching, smooth transition
and threshold autoregressive models. We adopt the latter approach, which is
characterised by a separation of the observations into dierent regimes based
on a threshold variable. Within each regime, the time series is then assumed
to be described by a linear model. In our multivariate context we use the
multivariate Tsay (1998) procedure to test for non-linearity in the data, using
the output gap or GDP growth as the threshold variable. In both cases the
test rejects linearity. Consequently, the threshold impulse responses (IR) are
generated using the general IR modelling introduced by Koop et al. (1996),
which allows for the non-linear propagation of shocks across regimes. Using
this framework, we can examine the sensitivity of economic activity to scal
policy shocks depending on the business cycle as well as of the size and the
direction of the shock.
6
In the baseline specication, we use the output gap as the threshold vari-
able
7
, as it divides economic development in phases of under- and overutilisa-
tion the two regimes under which we expect the eects of scal stimuli to
6
Tong rst proposed a threshold autoregressive (TAR) model in the late 1970s and rened
it during the early 1980s (Tong 1978, Tong and Lim 1980 and Tong 1983). Tsay (1998) and
Hansen (1996, 1997) made the threshold model applicable to a multivariate framework which
is now widely used in time-series analysis, as in Hansen (1999a, 1999b, 2000), Gonzalez and
Gonzalo (1997), Gonzalo and Montesinos (2000), Gonzalo and Pitarakis (2002), among
various others.
7
Koske and Pain (2008) demonstrate that the output gap plays an important role not
only for ex-post evaluation of policies but as well for real-time policy decisions. Furthermore
they argue that the output gap is a reliable real-time indicator in the short run.
3
dier.
8
To identify discretionary scal policy shocks in our non-linear model, we
employ structural identication following Blanchard and Perotti (2002) and
hence use exogenously determined elasticities for the working of automatic
stabilisers. However, we extend their approach and estimate the impulse re-
sponses based on time-varying elasticities. This allows us to identify discre-
tionary scal policy shocks even more reliably.
Our research shows that short-term scal multipliers in Germany are in
general moderate and that the state of the business cycle strongly matters for
the eects of scal policy shocks. In a linear benchmark model, the analysis
nds that the eect on GDP of reductions in tax and social security con-
tributions and of increased spending each corresponds to a short-term scal
multiplier of around 0.7, putting it in the range of other empirical results for
Germany. In addition the threshold model derives fundamentally new insights
on the eects of shocks, depending on their timing with respect to the business
cycle. Most importantly, scal spending multipliers are much larger in times of
a negative output gap but have only a very limited eect in times of a positive
output gap. Discretionary revenue policies, on the other hand, have generally
a more limited eect. With respect to the cycle, their impact is larger in the
upper than in the lower output gap regime. Various robustness checks, in-
cluding a dierent threshold specication, do not inuence the resulting policy
implications substantially.
The paper is organised as follows. Section 2 presents the main recent
empirical contributions in the literature, specically for Germany. Section 3
subsequently presents the empirical approach, and section 4 includes a detailed
description of the dataset. The empirical strategy for structural identication
and the estimation results are presented in section 5. Various robustness checks
are carried out in section 6, and section 7 concludes.
2 Empirical Literature
The literature on scal policy multipliers derives strongly diverging results
for the eects of scal stimuli on economic activity. The methods applied
in empirical analyses range from model simulations using dierent estimation
8
Within the robustness checks we pursue an analysis that employs a three-quarter moving
average of GDP growth itself as threshold.
4
and calibration techniques such as the IMF Multimod model, the OECD
Interlink or the ESCB New Area Wide Model - to reduced form equation
parameter estimation techniques. Surveys of the empirical literature, which
can be found in Hemming (2002), Spilimbergo et al. (2009) and Coenen et al.
(2010), demonstrate the great bandwith of spending and revenue multipliers.
Depending on the method and model specication, one-year scal spending
multipliers range between 0.2 and 2 for the US, while estimates for Germany
lie between -0.2 and 5.1.
9
While no consensus on the impact of scal policy on economic activity has
been reached, researchers generally agree on the importance of interdependen-
cies between scal and economic developments. Within the empirical literature
these interdependencies are most frequently analysed in vector autoregressive
(VAR) models. A focus in this literature lies on the identication of discre-
tionary scal policy shocks, for which most researchers rely on some form of
structural identication.
10
Most prominent are the recursive identication ap-
proach (Cholesky ordering), the sign-restrictions approach
11
and the structural
VAR approach using the identication procedure proposed by Blanchard and
Perotti (2002) (for a discussion on Cholesky ordering and sign-restrictions see
Perotti 2004).
In the present paper we follow Blanchard and Perotti (hereafter BP). BP
identify automatic stabilisers by incorporating exogenously given information
about the elasticities of revenues and expenditures with respect to GDP. In
their 2002 paper, they analyse the US economy between 1960 and 1997 and
nd that expansionary scal shocks increase output with a long-term scal
multiplier close to unity. Furthermore, they nd negative eects of tax and
spending increases on investment.
It is important to note that the results obtained from VAR studies depend
on the specic country analysed, as scal policy, the structure of the econ-
omy and the interplay of economic and scal developments dier substantially
9
However, the value of 5.1 is derived by Perotti (2006) only for public investment spend-
ing.
10
The narrative or event-study approach can also be used for identication. It identies
discretionary policy actions via specic historical events, such as contemporaneous press
reports, wars or war-related military spending, tax changes and elections. For further dis-
cussion see Ramey and Shapiro (1998), Edelberg et al. (1999), Eichenbaum and Fisher
(2005), Ramey (2006) and Favero and Giavazzi (2009) or Romer and Romer (2010).
11
For a current application to US data see Mountfort and Uhlig (2009), who nd the
highest multipliers for decit-nanced tax cuts.
5
across countries. Therefore, the analysis of shocks based on German data is
likely to yield dierent results than those for the US.
12
For Germany only a few
attempts analysing scal policy shocks are available, and all but one of them
employ the BP identication approach. The studies include H oppner (2001),
Perotti (2004), Heppke-Falk et al. (2010) and Bode et al. (2006). Afonso and
Sousa (2009) employ a recursive identication.
13
Hoppner (2001) uses quarterly cash data from 1970 to 2000 in a three-
variable VAR (government expenditures including government consumption,
investment and public transfers, such as subsidies; tax revenues from direct
and indirect taxes; and GDP). Based on his estimations, the eects of tax
shocks are by far larger than the eects of spending shocks. He estimates a
signicant scal multiplier for a tax increase of -1.59 in the long run (meaning
that a tax increase by one unit reduces GDP by more than 1.5 units), while
the multiplier for an expenditure shock equals only 0.23 (and is insignicant).
Perotti (2004) applies the BP approach to various countries, including Ger-
many, and analyses the eects of scal shocks on output, ination and the ten
year nominal interest rate. In his data set public spending is restricted to
public investment and consumption and does not include interest spending,
while the net revenue series is calculated by subtracting transfers from overall
revenues. Perotti argues that tax-nanced transfers have the reverse eects of
taxes and should therefore be substracted from overall taxes. For Germany
Perotti uses quarterly West German data from 1960 to 1989. Based on these
data denitions, he nds a short-term multiplier of only around 0.5 following a
positive shock of government spending, which fades out quickly after the rst
year.
Furthermore, Perotti identies a structural break in 1974 and re-estimates
the model in two subperiods (1960-1974 and 1975-1989). The structural break
inuences especially the eect of tax shocks. In the rst subsample he estimates
a short-term multiplier of 0.19 (after 4 quarters) for tax increases indicating
an expansionary eect of tax hikes - while in the second subsample the tax
multiplier for a positive shock is -0.03.
Heppke-Falk et al. (2010) follow the Perotti (2004) data denitions and
12
This is demonstrated by Burriel et al.(2010), who compare the eects of scal policy
in the US and the Euro area using the BP approach and nd a much higher persistency of
scal shocks in the US.
13
A detailed comparison of the rst three VAR studies for Germany can be found in Roos
(2007).
6
Table 1: German VAR Analysis, Impact of Fiscal Shocks on GDP
Sample short-run long-run
Eect of an increase in government spending on GDP
H oppner (2001) 1970-2000 positive insignicant
Perotti (2004)
a
1960-1989 positive insignicant/negative
Heppke-Falk et al. (2010) 1974-2008 positive* positive*
Bode et al. (2006) 1991-2005 positive insignicant
Afonso and Sousa (2009) 1980-2006 negative negative
Eect of a decrease in government revenues on GDP
H oppner (2001) 1970-2000 positive positive
Perotti (2004)
a
1960-1989 negative/insignicant insignicant
Heppke-Falk et al. (2010) 1974-2008 positive** insignicant
Bode et al. (2006) 1991-2005 positive insignicant
Afonso and Sousa (2009) 1980-2006 insignicant negative
* Signicant only for government investment; in sum with government consumption insignicant
** Signicant only for direct taxes; net revenue is insignicant
a : Two subsamples are tested, the rst 1960-1974, the second 1975-1989. If the results are dierent,
they are shown for the rst and the second subsample, respectively.
estimate a VAR covering a longer time series of quarterly German cash data
between 1974 and 2008. In their three variable specications (GDP, expendi-
ture and revenues), they nd a positive reaction of GDP to spending increases
that is signicant only in the contemporaneous quarter, with a spending multi-
plier of around one. Like Perottis rst subsample, they nd a positive reaction
of GDP to a revenue increase with a value of 0.12 in the quarter the shock
occurs.
Bode et al. (2006) estimate the structural three-variable VAR including
only pan-German data from 1991-2005. Based on a data denition similar to
that of Perotti (2004) and Heppke-Falk et al. (2010) they nd a signicant
positive eect of spending on GDP, with a one-year scal multiplier of around
0.5, as well as a signicantly negative but in size slightly smaller multiplier for
the eects of a revenue increase.
The latest available study for Germany is provided by Afonso and Sousa
(2009), who use a Cholesky decomposition for the structural identication in
a nine-variable VAR covering data from 1980 to 2006. They further add a
feedback rule in order to cover government debt dynamics, and nd a small
but signicant fall in GDP after a spending shock. The reaction of GDP to a
revenue increase is small but positive, supporting the nding by Perotti (2004)
for the rst subsample. They explain these results with a crowding-out
eect of public spending, but also a crowding-in eect of public revenues,
with both consumption and investment increasing after the shock as a result
of scal consolidation.
In summary, all the studies which use the BP identication scheme nd a
small positive scal multiplier for government spending increases, while Afonso
7
and Sousa (2009) estimate a small but negative eect based on a Cholesky
identication. With respect to tax cuts, the results of the discussed studies
dier strongly. Tax cuts increase GDP in the studies by H oppner (2001) and
Bode et al.(2006), while Heppke-Falk et al. (2010) and Afonso and Sousa
(2009) nd contractionary eects. Perottis (2004) results are sensitive to
the subsample analysed. Depending on the timespan covered, the impulse
responses display clear dierences and therefore indicate a variability of the
impact of scal shocks in dierent decades and macroeconomic environments.
Closely related to this last point, a drawback of all the empirical studies
presented is that they are bound to a linear estimation framework. That
is, they do not account for any asymmetry in the variable responses or the
relationship between the macro variables themselves. However, since they
provide a good starting point for a scal policy analysis, we use the linear
modelling as a benchmark and extend it to a non-linear threshold framework in
order to account for the possible asymmetries discussed in the introduction.
14
3 Methodology
Threshold VARs are piecewise linear models with dierent autoregressive
matrices in each regime. The regimes are determined by a transition vari-
able, which is either one of the endogenous variables or an exogenous variable
(Hansen 1996, 1997, Tsay 1998). In general it is possible to obtain more than
one critical threshold value and therefore more than two regimes, but for sim-
plicity we will focus on a model with two regimes only.
15
Let a set of k stationary endogenous variables with y
t
= (y
1t
, ..., y
kt
)

and
T observations describe a VAR of nite order p
y
t
=
0
+
1
y
t1
+ ... +
p
y
tp
+ u
t
, (1)
where
0
is a k-dimensional vector containing deterministic terms such
as a constant, a linear time trend or dummy variables.
i
with i = 1, ..., p
are squared coecient matrices of order k, and u
t
is a sequence of serially
uncorrelated random vectors with mean zero and covariance matrix Cov(u
t
) =

u
. We can rewrite equation (1) in the compact form
y
t
= X
t
+ u
t
, (2)
14
An interesting non-linear analysis of German scal policy using a dierent methodology
is Hoppner and Wesche (2000), who apply a Markov-switching approach to scal policy
eects in Germany and nd time-varying eects.
15
The two-regime setup is also best for our scal policy analysis over the business cycle
since the general concept of the business cycle is based on a distinction between an upper
(positive output gap) and a lower (negative output gap) regime.
8
with = (
0
,
1
, ...,
p
) and X
t
= (1, y
t1
, ..., y
tp
)

. Following this nota-

tion, a threshold VAR is represented by
y
t
=
1
X
t
+
2
X
t
I[z
td
z

] + u
t
. (3)
z
td
is the threshold variable determining the prevailing regime of the sys-
tem, with a possible lag d. I[] is an indicator function that equals 1 if the
threshold variable z
td
is above the threshold value z

and 0 otherwise. The

coecient matrices
1
and
2
, as well as the contemporaneous error matrix
u
t
are allowed to vary across regimes. The delay lag d and critical threshold
value z

are unknown parameters and determined alongside the parameters.

In the linear and the non-linear model we face the problem that the con-
temporaneous errors are not uncorrelated with each other, i.e.
u
is not a
diagonal matrix. In this case scal policy shocks are not identied, since the
correlation of the error terms indicate that a shock in one variable is likely
to be accompanied by a shock in another variable. Following Blanchard and
Perotti (2002), we identify the policy shocks using an AB model for structural
identication in the error-covariance matrix. The linear model thus becomes
16
Ay
t
= CX
t
+ B
t
, (4)
assuming that u
t
= A
1
B
t
where
u
= A
1
BB

A
1

and B is a k k
matrix of parameters.
17
The non-linear model can be correspondingly written as
A
n
y
t
= C
1
X
t
+ C
2
X
t
I[z
td
z

] + B
n

n
t
, (6)
where A
n
and B
n
dier from A and B in the linear model, since they are
based on the regime-dependent errors. As before,
i
= A
1
i
C
i
, and Cov(u
n
t
) =

n
= A
1
n
B
n
B

n
A
1
n
. The exact identication procedure is explained in section
5.3.
Before estimating a non-linear model we need to test if the system is indeed
non-linear. Following the testing approach developed by Tsay (1989, 1998) we
rst identify a series z representing the threshold variable with = z
0
<
z
1
< ... < z
s1
< . z needs to be stationary with a continuous distribution,
restricted to a bounded set S = [z, z], where S is an interval on the full sample
16
A detailed description and derivation of the AB model, as well as the corresponding
A and B models can be found in Amisano and Giannini (1997), L utkepohl (2005) and
L utkepohl and Kr atzig (2004). Further applications of the AB model can be found in Pagan
(1995), Breitung and L utkepohl (2004) and Blanchard and Perotti (2002).
17
Alternatively, the AB model can be represented in the error component form
Au
t
= B
t
. (5)
9
range of the threshold variable. The interval should be trimmed in order to
assure a minimum number of observations in each subsample.
The lag order p and the threshold lag d need to be determined a priori,
which in case of p is achieved by applying the normal information criteria in the
linear VAR estimation. For the choice of d we will rely on economic reasoning.
The regression framework of equation (2) can be rewritten as
y

t
= X

t
+ u

t
, t = h + 1, ...n , (7)
where, as before, denotes the parameter matrix, X
t
= (1, y

t1
, ..., y

tp
)

,
and h = max(p, d).
We reorder the cases according to the threshold variable z
td
, denoting the
i-th smallest element of the interval S as z(i) (equals the m-th smallest value
of all observations.
18
) The arranged regression can be written in the form
y

t(i)+d
= X

t(i)+d
+ u

t(i)+d
, i = 1, ..., n h (8)
where t(i) is the time index of z(i). In short, we order the values of the
threshold variable according to its size and split the sample according to the
threshold value z(i). The model is estimated with the m observations below
z(i) by OLS to obtain

m
. Subsequently, OLS is performed again for the
rst m + 1 observations with z(i + 1) and so on. The result is a sequence of
OLS regressions, each using the rst m ranked observations. For each of these
regressions, we keep the one-step ahead predictive and standardised residuals
and , calculated with

t(m+1)+d
= y
t(m+1)+d

m
X
t(m+1)+d
(9)

t(m+1)+d
=

t(m+1)+d
[1 + X

t(m+1)+d
(
m
i=1
X
t(i)+d
X

t(i)+d
)
1
X
t(m+1)+d
]
1/2
. (10)
In order to analyse threshold behaviour we test for white noise in the re-
gression

t(l)+d
= X

t(l)+d
+ w

t(l)+d
, l = m
0
+ 1, ..., n h (11)
where m
0
denotes the starting point of the recursive least squares estima-
tion. If = 0 the data is generated by a linear model.
19
Consequently, we
18
Remember that the interval S = [z

, z] is trimmed. The trimming percentage is the

percentage of observations of the whole sample below m
0
. m
0
corresponds to z(i) with
i = 1.
19
The sequential OLS estimates are consistent estimates of the lower regime parameters
as long as the last observation used in the regression does not belong to the upper regime.
In this case, the predictive residuals are orthogonal to the corresponding regressor and y
t
is
linear. However, if y
t
follows a threshold model, the predictive residuals will not be white
noise and correlated with X
t(i+d)
. As a consequence the least squares estimator would be
biased.
10
test the hypothesis H
0
: = 0 against the alternative H
1
: = 0. Tsay (1998)
proposes the following test statistic:
C = [n h m
0
(kp + 1)]{ln[det(S
0
)] ln[det(S
1
)]} (12)
with det() being the determinants of
S
0
=
1
n h m
0
nh

l=m
0
+1

t(l)+d

t(l)+d
(13)
and
S
1
=
1
n h m
0
nh

l=m
0
+1
w
t(l)+d
w

t(l)+d
. (14)
The test statistic is asymptotically chi-square with k(pk + 1) degrees of
freedom. If the test detects a threshold in the DGP, the coecients can be
estimated conditional on a sum of least square minimisation over both regimes.
That is, for a given value of z, the LS estimate of
(i)
for regimes (i) = 1, 2 is

(i)
(z) = (
(i)

t
X
t
(z)X
t
(z)

)
1
(
(i)

t
X
t
(z))y
t
(15)
with residuals u
t(i)
(z) = y
t
X
t
(z)

(i)
(z), and residual variance

2
(i)
(z) =

(i)
t
u
2
t(i)
(z)
n
(i)
k
, (16)
where

(i)
t
is the sum of all observations in regime (i) and n
(i)
is the number
of observations in regime (i). The sum of squared residuals is

R(z) =

R
(1)
(z) +

R
(2)
(z) , (17)
where

R
(i)
(z) = (n
(i)
k)
2
(i)
(z). Finally, the conditional threshold value
z is obtained by
z = argmin
z

R(z) . (18)
In the case of a test result that suggests a threshold eect, we wish to
apply an impulse response (IR) analysis that is able to capture non-linearities.
Gallant et al. (1993), Koop (1996) and Koop et al. (1996) point out that
in non-linear models the eect of a shock depends on the entire history of
the system up to the point when the shock occurs. Thus, it is necessary to
model the IRF conditional on this history, and as a consequence conditional
11
on the size and the direction (sign) of the shock.
20
For this purpose we cannot
apply linear IR functions, as they are history-independent, i.e. they do not
depend on a particular history of the data up to time t. They are symmetric
in the sense that a shock of
tm
has exactly the opposite eect of a shock
of size +
tm
and they are linear as they are proportional to the size of the
shock. Hence, they cannot be applied here. Instead, we will model generalised
impulse response functions (GIRF) introduced by Koop et al. (1996), which
address these problems and which are applicable to both linear and non-linear
models.
Dening
t
as a shock of a specic size, m as the forecasting horizon and

t1
as the history or information set at time t 1, Koop et al. (1996) dene
the GIRF as the dierence between two conditional expectations with a single
exogenous shock:
GIRF = E[X
t+m
|
t
,
t+1
= 0, ...,
t+m
= 0,
t1
]
E[X
t+m
|
t
= 0,
t+1
= 0, ...,
t+m
= 0,
t1
] . (19)
The GIRF allows the regimes to switch after a shock, a characteristic that
is responsible for the dierent outcomes of positive and negative shocks as well
as their size.
21
. We are thus able to relax the assumption that shocks occur-
ring in a recession are just as persistent as shocks occurring in an expansion.
The calculation of the GIRF induces some computational eort and the exact
algorithm is described in Appendix (A).
4 The Data
To keep the analysis as parsimonious as possible, we include only three vari-
ables: government spending, government revenue and GDP.
22
In the non-linear
specication, we need an additional threshold variable in order to distinguish
between good and bad economic times. Here we rely on the output gap as
an indicator for the dierent phases of the economic cycle, which is generally
20
In fact, non-linear time series models do not have a Wold representation and the as-
sumption that no shocks occur in intermediate periods may give rise to misleading inferences
concerning the propagation mechanism of the model.
21
GIRF were applied in several empirical applications, for example in Balke (2000),
Atanasova (2003), Root and Lien (2003), J aaskela (2007) A detailed description of GIRF
for the univariate case can also be found in Potter (2000).
22
A ve-variable model that contains investment and consumption might allow us to study
the transition of policy shocks in the economy in more detail, but estimating a ve-variable
model is impractical because the number of coecients in the linearity test and the TVAR
rises in proportion to the number of coecients in the standard linear model. This aects the
size and power of the tests. Therefore, the present paper will restrict itself to a three-variable
specication and leaves the proposed extension to future work.
12
seen not only as a reliable ex-post but also as a reliable real-time indicator for
policy-makers (see Koske and Pain 2008).
The data is compiled from the Deutsche Bundesbanks national accounts
database and dened according to the European System of National Accounts
(ESA) 1979 and 1995. The advantage the national accounts data has over cash
data is that its data are adjusted for special events and distortions caused, for
example, by lagged payments of taxes.
23
Additionally, we remove the eect of
the liquidation of the German Treuhand in 1995 (199.6 billion Euro in total)
and revenues from the auction of the UMTS licenses in 2000 (50.8 billion Euro).
The dataset is quarterly and covers the period from the rst quarter of
1976 to the fourth quarter of 2009, giving 136 observations. Generally we
would have been able to start our analysis in the rst quarter of 1970, but
we decided to exclude the rst ve years in order to avoid a structural break
due to a policy shift after 1975.
24
The structural break at reunication is
eliminated by prolonging the series for reunied Germany backwards with
West German growth rates. Our estimations are based on data in real terms
(all three variables are deated by the GDP deator with a value of 1 in the
year 2000) and seasonally adjusted by applying the BV 4.1 procedure of the
German Federal Statistical Oce.
25
The output gap is calculated with the Hodrick-Prescott lter ( = 1600)
applied to the real GDP series. To avoid a distortion of the results at the lower
and upper bound, we prolong the series with its own linear trend in the past
(1960-1970) and the future (2009-2019). The real output gap variable is then
calculated as the dierence between actual real GDP and potential real GDP
(measured by the HP-ltered trend) as a percentage of potential GDP.
The scal series oer a great variety of possible compositions. Seminal stud-
ies such as BP (2002) and Perotti (2004) dene public spending very narrowly
as government investment plus government consumption, and public revenues
as general government revenues (excluding social security) minus transfers. Al-
though many papers follow this denition (see the discussion in section 2), we
argue that it is not well-suited for an analysis of scal policy in Germany, since
social insurance accounts on average for more than 40% of total revenues and
23
See ECB (2007) for a denition of government nance statistics according to ESA and
standard methods of national accounting.
24
Before 1976 the German government - inspired by Keynesian macro-economics - aimed
at an active stabilisation of the business cycles via frequent temporary tax and expenditure
measures. The most important measure aimed at economic stimulation was the introduction
of an investment bonus of 7.5% for all investment in machinery and equipment realised
between 1 December 1974 and 30 June 1975. These measures contributed to a high volatility
of tax revenues and spending, which is reected especially in the growth rates of the series,
with a breakpoint identied by Perotti (2004), for instance, to be 1974:4.
25
A discussion of the methodology applied in the Berliner Verfahren can be found at
http://www.uni-mannheim.de/edz/pdf/eurostat/06/KS-DT-06-012-EN.pdf.
13
Table 2: Unit Root Tests
ADF Phillips-Perron
Lags (SIC) t-statistic p-value t-statistic p-value
Revenues 7 -4.274 0.0007 -8.984 0
Expenditures 3 -5.388 0 -11.721 0
GDP 1 -4.914 0.0001 -6.999 0
Output Gap 9 -5.155 0 -3.705 0.005
H
0
: series has a unit root.
for a large part of overall public spending. Furthermore, economic stimulation
is often explicitly pursued via the social security system. For example, during
the 2008/2009 recession large parts of the German scal stimulus were imple-
mented through decit-nanced cuts in social security contributions. Even if
considered as a pure redistribution (such as in Perotti 2004, Heppke-Falk et
al. 2010, or Bode et al. 2006), the social security system can have far-reaching
eects on private consumption based on dierences in the savings rate of net
payers and net recipients. These consumption eects, in turn, can inuence
overall growth. Thus, we include social insurance in our analysis.
Unemployment spending is, due to its strong dependence on the business
cycle, subtracted from the expenditure side and enters the revenue series with
a negative sign in order to satisfy the precondition of the BP structural identi-
cation approach that automatic stabilisers only apply on the revenue side. The
treatment of public interest spending diers over the existing studies. Perotti
(2004) and Fern andez and Cos (2006) argue that interest spending is not part
of discretionary scal policy and should be excluded from the data series, while
Blanchard and Perotti (2002) argue that interest payments should be included
as they reect a normal transfer of resources from the public to the private
sector (and thereby inuence economic growth). We follow the approach of
Perotti (2004) and subtract interest payments from the expenditure and the
revenue variable.
26
Taken together, this leaves us with a public spending series dened as total
current public spending excluding net interest (i.e. interest spending minus
dividends received by the government) and unemployment insurance spending.
Our revenue series includes social security contributions but is diminished by
net interest spending and unemployment insurance spending.
Except for the output gap (which is stationary), we apply the logarithm
to all series and take the rst dierences in order to achieve stationarity. The
resulting quarter-to-quarter growth rate series and the output gap are plotted
in gure 1. All series tend to revert to their mean.
26
This approach is in line with other studies for Germany, such as Heppke-Falk et al.
(2010) and Bode et al. (2006).
14
Figure 1: Growth Rates of Series
-.02
-.01
.00
.01
.02
.03
80 85 90 95 00 05
Expenditures
-.04
-.03
-.02
-.01
.00
.01
.02
.03
80 85 90 95 00 05
GDP
-.03
-.02
-.01
.00
.01
.02
.03
.04
80 85 90 95 00 05
Revenues
-.03
-.02
-.01
.00
.01
.02
.03
.04
80 85 90 95 00 05
Output Gap
15
Table 3: Descriptive Statistics
Variable Mean Maximum Minimum Std. Dev.
Revenues 0.0049 0.031 -0.026 0.012
Expenditures 0.0039 0.021 -0.013 0.007
GDP 0.0048 0.023 -0.033 0.007
Output Gap -0.0001 0.036 -0.029 0.014
We also employ the standard unit-root methodology, i.e. the Augmented
Dickey-Fuller (ADF) and Phillips-Perron test. It is necessary to choose the
number of augmentation lags to account for serial correlation in the Dickey-
Fuller regressions, for which we use the Schwarz Information Criterion (SIC).
For all four series, both tests include a constant but no trend. Table 2 shows the
results for the growth rates. The values indicate that the series are stationary
by rejecting the null hypothesis of the existence of a unit root. Additionally,
descriptive statistics of the three series are shown in table 3.
5 Estimation
5.1 Model Specication
The VAR of equation 1 consists of a three-dimensional system of endoge-
nous variables y
t
= [T
t
, G
t
, GDP
t
], with T
t
, G
t
and GDP
t
being the growth
rates in government revenues, government spending and GDP, respectively. A
constant is included in y
t
.
For the optimal lag length we conduct various model selection tests, which
provide dierent lag order suggestions. While the Schwarz Information Crite-
rion (SIC) suggests the use of only one lag, the Akaike Information Criterion
(AIC) and the Hannan-Quinn (HQ) Criterion suggest the use of four lags
and the prediction error the use of six. Although the majority of the criteria
propose a higher order, we follow the SIC and specify the benchmark speci-
cation with one lag. We base this choice on the same reason for using only
three variables: the high cost of estimating additional parameters and there-
fore of over-tting in the non-linear model (every additional parameter added
decreases the power of the estimation substantially; see for example Hansen
(1996) for a Monte Carlo proof).
Using one lag in the VAR, the Breusch-Godfrey Lagrange multiplier test
for serial correlation does not reject the null hypothesis of no serial correlation
for the tested lag numbers 2 and 5 with p-values of 0.115 and 0.218, respec-
tively. On the other hand, the Portmanteau test does reject the null at least
at the 5% level for higher lag values.
27
It is thus important to check the model
27
However, the same is true if the model is estimated with four and six lags. Therefore
we do not base the choice of the lag order on the autocorrelation properties.
16
Table 4: Tsay Threshold Test
d Test statistic p-value
a
Threshold value
0 19.971 0.0676 -0.001516
1 39.916 0.0000 -0.001510
2 32.138 0.0013 -0.001516
H
0
: linearity, H
1
: threshold behavior
d: lags in the threshold variable
estimation properties in a profound robustness analysis. Furthermore, since
the standard errors might be underestimated, we have to be careful in inter-
preting the condence regions. To rule out that the policy implications we will
nd rely to a great part on imprecise structural identication, we will test the
eects of alternative values for a
1
as well as changes in the general identica-
tion procedure through the application of the Cholesky decomposition in the
robustness checks.
5.2 Threshold Tests
The test results for the null of linearity with one lag in the VAR and
dierent lags in the output gap are presented in table 4. The Tsay test statistic
is computed using a 30% trimming percentage and the test rejects linearity
of the system for all three threshold specications. We will continue to use
one lag in the threshold variable, in order to account for moderate economic
rigidities.
The estimated threshold value for a specication with one lag in the VAR
and one lag in the threshold variable is -0.0015. Being in close neighbourhood
to zero, this value justies the classication of the two regimes as representing
periods of output above and below its potential level. The two regimes are
shown in gure 2. The sample splits into 45 observations in the lower regime
and 91 observations in the upper regime, with a total of 15 regime switches.
With two of those being of very short length, this gives us approximately six
complete business cycles within 39 years.
5.3 Identication
As described earlier, we follow the identication procedure developed by
Blanchard and Perotti (2002). In the SVAR representation Au
t
= B
t
, u
t
=
(gdp
t
, g
t
, t
t
) is the vector of reduced-form error terms for the GDP, government
spending and revenue equation, respectively. The vector of structural shocks
is given with
t
= (
GDP
t
,
T
t
,
G
t
) with Cov(
t
) = I
3
and
GDP
t
,
G
t
and
T
t
corresponding to the GDP, tax and spending shocks. After estimating the
reduced form VAR, we can use the reduced-form residuals u
t
to determine the
elements of A and B. But prior to that, some identifying assumptions need to
17
Figure 2: Output Gap Regimes, Threshold Value=0.15%
-.006
-.004
-.002
.000
.002
.004
.006
1980 1985 1990 1995 2000 2005
OUTGAP THRESH
be made. First, the innovation in the scal variables g
t
and t
t
can be described
as a linear combination of three types of shocks, (i) the automatic response
of government expenditure and revenue to real output, (ii) the systematic,
discretionary response of expenditure to shocks in revenue and of revenue to
shocks in expenditure and (iii) the random, discretionary scal policy shocks,
which are the underlying structural shocks
G
t
and
T
t
to be identied. We
also think of unexpected changes in GDP (gdp
t
) as a function of shocks in
government spending, revenue and a structural shock in GDP itself. With
these assumptions, we can write:
t
t
= a
1

GDP
t
+ a
2

G
t
+
T
t
g
t
= b
1

GDP
t
+ b
2

T
t
+
G
t
(20)
gdp
t
= c
1

T
t
+ c
2

G
t
+
GDP
t
.
We can rearrange this system to reconstruct the AB representation Au
t
=
B
t
, with
A =

1 0 a
1
0 1 b
1
c
1
c
2
1

and B =

1 a
2
0
b
2
1 0
0 0 1

. (21)
Using information about the tax and transfer system to determine the
coecients in A and B, Blanchard and Perotti apply the following procedure:
I) In the rst step institutional information on the German public nance
system is used in order to identify the coecients a
1
and b
1
. We have to
consider that the two coecients incorporate two distinct eects of activity
on spending and taxes. They capture the automatic stabilisers, which are the
automatic eects of economic activity on the scal variables under existing
18
0.95
0.96
0.97
0.98
0.99
1.00
1.01
1.02
1.03
1970 1975 1980 1985 1990 1995 2000 2005
Figure 3: Time-Varying Elas-
ticities
-.04
-.03
-.02
-.01
.00
.01
.02
.03
.04
.05
-.03 -.02 -.01 .00 .01 .02 .03 .04
Output Gap
A
d
ju
s
t
e
d

R
e
v
e
n
u
e
s
Figure 4: Adjusted Revenue - Output
Gap Correlation
scal institutions. In addition, they capture any discretionary adjustment
of scal policy to unexpected exogenous changes in economic activity within
the same quarter. As long as we assume that it takes scal policy to react
some time to changes in GDP due to democratic, legislative and bureaucratic
processes in decision making and implementation, the use of quarterly data
basically eliminates the second channel. It is thus valid to assume that a
1
and b
1
solely capture the automatic responses of scal variables to GDP. They
are calculated using the OECD framework by Girouard and Andre (2005), the
dierence being that we use quarterly instead of annual data. For the aggregate
elasticity of N tax series with respect to output Girouard and Andre (2005)
apply the following formula:
a
1
=
N

i=1

T
i
,B
i

B
i
GDP

T
i

T
(22)
where

T are the net taxes, with

T =

N
i

T
i
and

T
i
being taxes of type
i, which take on a positive value for taxes and a negative value for transfers.

T
i
,B
i
denotes the elasticity of tax i with respect to its tax base B
i
and
B
i
GDP
denotes the elasticity of the tax base to GDP. The exact calculation of the
elasticity a
1
and the taxes and tax bases in use are described in Appendix (C).
Following the OECD approach, we nd an elasticity around 1, which lies in
the middle of elasticities applied in studies relying on similar data.
28
However, one could object that the shares of the dierent revenue and
expenditure components in net revenues vary strongly over time, which we
demonstrate in gure 3. Even if the elasticities of the subcomponents were
stable - this would make the application of time-varying elasticities necessary.
In this respect we extend the basic BP approach and use the time-varying
elasticities instead.
28
We also estimate the model based on elasticities of 0.5 and 1.5 to test the robustness of
our results; see section 5.1.
19
Additionally, we test the viability of the applied elasticities through corre-
lation between the output gap and the revenue series, which we adjusted for
the automatic responses based on the calculated elasticity. A non-zero corre-
lation means that the elasticities are misspecied and discretionary shocks not
precisely identied. Figure 4 shows the regression result and the scatter-plot
of the output gap on the x-axis and the adjusted growth rates of revenues on
the y-axis. We detect no correlation between the two series, which speaks in
favour of the elasticity applied and indicates that non-linearity is more likely to
be rooted in discretionary scal policy reactions than in automatic stabilisers.
The identication of b
1
is easier. It can be set to zero, as the main com-
ponent of primary government spending (unemployment transfers) is included
in net revenues.
II) In the second step, we construct the contemporaneous inuence of rev-
enues and expenditure on GDP, c
1
and c
2
. With the estimates of a
1
and b
1
the cyclically adjusted reduced-form scal policy shocks (revenue and spend-
ing residuals) are calculated with t

t
= t
t
a
1
gdp
t
and g

t
= g
t
b
1
gdp
t
= g
t
.
These can be used as instrument variables in the third equation of system (16).
They are considered as instruments since they are no longer correlated with

GDP
t
(though still correlated with each other). Therefore, we can consistently
estimate the coecients c
1
and c
2
with least squares estimation.
III) In the nal step, the remaining parameters a
2
and b
2
need to be deter-
mined. In the literature it is controversially discussed whether taxation follows
spending (b
2
= 0) or spending follows taxation (a
2
= 0) (see e.g. Kollias and
Paleologou 2006, Hoover and Sherin 1992 or Koren and Stiassny 1998). In the
baseline model, a
2
is constrained to zero and b
2
is estimated (revenue decisions
come rst).
The time-varying elasticity of a
1
with a mean around one and the described
identication yield the following matrices of contemporaneous relationships
R
lin
, and R
nonlin
for the linear and non-linear model:
29
R
lin
=

1 0 1.0119
0.0840 1 0
0.2070 0.2744 1

R
nonlin
=

1 0 1.0119
0.0964 1 0
0.1881 0.2293 1

.
(23)
We see large dierences between the non-linear and the linear model. For
example, the contemporaneous inuence of revenue and spending shocks on
29
As a simplication (and to save space) the resulting A and B matrices will be presented
combined and are given in all following sections in the form
R
lin
=

1 0 a
1
b
2
1 0
c
1
c
2
1

.
20
unexpected changes in GDP (lower left and middle entry) are substantially
smaller in the non-linear case.
5.4 Impulse Response Analysis
In the following subsections we will present and discuss our estimation
results based on impulse response functions (IRF). We start with the linear
benchmark model and then discuss the IRFs for scal shocks in the lower and
the upper regime of the threshold model. Throughout the GIRF generation,
we update the output gap after each forecasted quarter using a one-sided HP-
lter. Additionally we review the eects of an increase in the size of shocks
and the dependence of the results on the denition of the threshold variable
(the GDP growth rate series is used as an alternative).
5.4.1 Linear Impulse Response
The linear impulse responses for a one-time shock in revenues and spending
are presented in gure 5. Since the purpose of this paper is to analyse the
impact of scal policy on GDP (not vice versa), we do not show the impulse
responses to a shock in GDP. As a benchmark we apply a shock of 2%.
We nd that government spending reacts weakly but positively to a revenue
shock, with an IRF that returns to zero within two periods. Since we have
set the contemporaneous reaction of revenue to a public spending shock to
zero (see section 5.3), revenues react with a lag of one period. The response is
negative for this period and zero thereafter.
The lower two gures show the response of GDP growth. The impact
of revenue increases on GDP is small and negative, with a contemporaneous
eect of -0.3%. The positive spending shock has a small positive impact on
GDP, with a contemporaneous value that is slightly larger than the absolute
value for revenue changes, and a cumulative eect of about 0.35% after three
quarters. Taking into account that, over the observation period, government
spending and revenues equal on average 41% and 39% of GDP, respectively,
we obtain a scal spending multiplier of 0.7 and a revenue multiplier of -
0.66 (all multiplier results are presented in table 5). The scal multipliers for
expenditure and revenue policies are of similar size and are generally moderate
- meaning that a stimulus of 1% of GDP increases GDP in the short run by
substantially less than 1%. This would indicate that public spending causes a
partial crowding out of private activity - a result in line with the ndings of
comparable SVAR studies using the Blanchard-Perotti identication.
In the following we present the GIRFs. In order to directly compare positive
and negative shocks, the linear IRFs are included and the negative impulse
responses are shown mirror inverted.
21
Figure 5: Linear Impulse Responses, Shocks in R and G
5.4.2 Lower Regime, 2% Growth Shock
The GIRFs for a 2% scal shock are shown in gure 6. The red (evenly
dotted) IRs represent the linear model, while the solid and variable-dotted lines
show the responses to positive and negative shocks, respectively. Foremost, we
nd clear dierences between the lower output gap regime and the linear model,
especially in response to spending shocks. As such, the GIRF for a spending
shock on revenues is negative after one period, but becomes positive for the
second quarter after a shock and again negative after the fourth quarter.
The lower right gure reveals that the GDP response to a scal spending
shock in periods of negative output gaps is larger and more persistent than the
linear model suggests. Although we nd a lower contemporaneous inuence
of spending on GDP than in the linear identication (compare system 23),
the cumulative response is larger and more persistent. In specic, the scal
multiplier increases to 1.04 four quarters after the shock and is still 0.99 ten
quarters after the shock (see table 5). Thus, with almost no crowding-out in the
long run, the spending multiplier slightly above unity indicates the possibility
for scal policy to stimulate unused factors of production. The result further
implies that the linear model underestimates especially the short-run impact
of government spending activity under negative output gaps.
Comparing this result to the revenue multipliers, we nd only a small
22
Figure 6: Lower Regime: 2% Growth Shock
dierence between the linear and non-linear model. The cumulative short-
run revenue multiplier decreases to 0.5 in absolute terms, compared to 0.66 in
the linear model, indicating that tax reductions do appear to be less well-suited
to pushing the German economy out of a recession than expenditure increases.
On the other hand, tax increases in a period of a negative output gap do seem
to harm the economy especially in the short-run less than expenditure cuts.
Figure 6 further shows that dierences between the positive and negative
GIRFs are relatively small, with the reason being that the output gap responds
only sluggishly to economic growth. As an example, assume that in the lower
regime a positive spending shock on GDP pushes the economy into the upper
regime, while a negative shock does not. With dierent parameter estimates
for the two regimes we would expect dierent responses. But since the output
gap is very persistent, a regime change does not occur frequently at a small
shock size and the positive and negative responses can be very similar.
5.4.3 Upper Regime, 2% Growth Shock
For the upper regime - reecting the periods when the economy is above
potential output - the GIRFs for a 2% growth shock are shown in gure 7.
While the responses to revenue shocks are again close to the linear model
IRFs (and the lower regime), the GIRF of revenues to a positive spending
23
Figure 7: Upper Regime: 2% Growth Shock
shock is now negative for at least eight quarters following the shock.
The most striking dierence to the lower regime is the response of GDP
to a spending shock. The contemporaneous response is small and it becomes
negative from the rst quarter following the shock. As a consequence, the
scal multiplier, at 0.36 after four quarters, is substantially lower than in the
linear model and the lower regime values. This smaller multiplier indicates a
substantial crowding-out of private activity, even in the short-run. Thus, our
model suggests that governments should refrain from expansionary scal policy
through spending increases in periods where a positive output gap prevails.
The upper regime revenue multipliers are comparable to the lower regime,
with -0.58 and -0.53 after four and ten quarters, respectively. Based on the
observations that spending multipliers in the upper regime are substantially
smaller than the lower regime ones, with an included and large crowding out
eect, it seems more eective to employ spending policies only under a negative
output gap regime, and limit tax policies to the times when output is above
its potential.
5.4.4 Comparison Lower and Upper Regime, Increasing Shock Size
In general, the size of the shock can lead to noticeable dierences in the
responses of the GIRFs, even with the sluggishness of the output gap. Figures
24
Table 5: Fiscal Multipliers
4-Quarter
spending shock revenue shock
Size 2% 5% 2% 5%
Sign pos neg pos neg pos neg pos neg
Linear model 0.7 -0.66
Lower regime 1.04 -0.86 1.27 -0.84 -0.5 0.51 -0.48 0.53
Upper regime 0.36 -0.60 0.26 -0.84 -0.58 0.61 -0.60 0.62
10 Quarter
spending shock revenue shock
Size 2% 5% 2% 5%
Sign pos neg pos neg pos neg pos neg
Linear model 0.69 -0.68
Lower regime 0.99 -0.84 1.28 -0.83 -0.49 0.49 -0.47 0.51
Upper regime 0.34 -0.56 0.28 -0.75 -0.53 0.54 -0.54 0.57
Calculated based on ratio of spending and revenue to GDP.
8 and 9 show that, while the responses of revenue shocks are almost identical
to the small shock size results, especially the upper regime GIRFs following
expenditure shocks change noticeably, with increased dierences between pos-
itive and negative responses. Accordingly, the scal spending and revenue
multipliers provided in table 5 change substantially only for larger expendi-
ture shocks: The short term multiplier of a 5% spending increase is 1.27 in the
lower but only 0.26 in the upper regime. Spending reductions of 5% have in
both regimes a short-term multiplier of -0.84.
6 Robustness Checks
To make sure that our results are robust and reliable, we test the inuence
of the application of an alternative threshold variable, of alternative structural
identication schemes, variations in the exogenous elasticity, the data sample
and the threshold value.
6.1 GDP Growth Threshold
An alternative threshold variable is GDP growth. By using growth rates
we analyse how the eect of scal shocks diers if GDP growth is below or
above a certain threshold rate. Since GDP growth is relatively volatile, the
threshold series is dened as the three-quarter moving average of the series.
Furthermore, in order to account for economic rigidities the threshold series
follows the variables with one lag. The Tsay test rejects linearity and we obtain
a threshold value of 0.0035 (real GDP growth of 0.35%), spitting the sample
25
into 54 observations in the lower, and 82 observations in the upper growth
regime. The responses for a 2% growth shock are presented in gures 10 and
11.
In general, most of the responses change moderately, with the clearest
changes observed in the responses of the scal variables to one another. How-
ever, the implications we derived in the baseline specication do not change
signicantly. The linear model underestimates the scal spending multipliers
in the lower, and overestimates them in the upper regime, even though this
eect is smaller with GDP growth as the threshold variable. The results for a
revenue shock on GDP do not show drastic changes, although the revenue mul-
tiplier in the upper regime is somewhat smaller than the lower regime value.
Thus, using a dierent measure for economic performance as threshold variable
has almost no impact on the estimation.
6.2 Structural Identication
We employ dierent (xed) values for a
1
in the structural identication,
accounting for diverging values in the literature. In the identication of section
5.3, we allowed the elasticity to be time-varying for a less biased structural
identication, with a mean of a
1
to be around 1, whereas values in the empirical
literature range from 0.46 as in Bode et al. (2006) to above one as in H oppner
(2001) and Leibfritz (1999). In order to rule out any impact of the specic value
of the calculated elasticity on the implications, the IRFs are estimated for two
alternative elasticities, 0.5 and 1.5. Figures 12 and 13 show the resulting linear
IRFs and GIRFs. The only noticeable dierence to the benchmark model is
the magnitude of the response of GDP to a revenue shock, which increases
(decreases) substantially in size for an elasticity of 0.5 (1.5) for both the linear
and non-linear model. At any rate neither the implications for the threshold
model in response to a revenue shock, nor those for the model in response to
a spending shock change with dierent elasticities; we can therefore conclude
that the model is robust to changes in a
1
.
In a second robustness check we apply the Cholesky decomposition in order
to determine the extent to which the identication approach matters. We
compare the IRFs for the alternative variable orders GDP R G and R
G GDP, shown in gures 14 and 15 for the lower growth regime (including
the linear model), in gures 16 and 17 for the upper regime (and a shock size of
2 SE, which roughly corresponds to a 2% revenue and 1.5% expenditure shock).
For both impulse orders the results of the GDP responses change drastically,
especially in the linear model (the responses in the scal variables are only
mildly aected). In the linear model, for both impulse orders, the response
of GDP to a revenue shock is positive, albeit small. This result is very close
to the one found by Afonso and Sousa (2009), who also apply a Cholesky
identication. The linear IR of GDP to a spending shock is very sensitive to
26
the change in the impulse order. Being entirely negative for GDP R G,
it accumulates to a positive multiplier for R G GDP. On the other hand,
the threshold specication shows that the response of GDP to a spending shock
is robust in the impulse ordering (although we nd the same positive impact
of a revenue shock). In the lower regime, the spending multipliers are similar
to those obtained with the Blanchard and Perotti identication. In the upper
regime multipliers do not change signicantly for the order R G GDP,
but decrease drastically for the alternative. However, in both cases, the upper
regime responses yield signicantly smaller scal spending multipliers than the
lower regime.
This analysis leads to two conclusions. First, the exact structural iden-
tication is of great importance, for the non-linear model but even more for
the linear specication. Since the Blanchard and Perotti (2002) identication
approach focuses mainly on the interaction between revenues and GDP, it is
not surprising that the Cholesky decomposition changes the GDP response to
a revenue shock in the linear and the non-linear model (and for both variable
orderings). Second, we see that the threshold model is more robust to changes
in the identication strategy than the linear model. The comparison of the two
regimes provides more room for interpretation than the volatility-prone linear
model allows. That is, the implications from the non-linear estimation remain
very similar. In the lower regime, we observe higher absolute scal spending
and revenue multipliers, in the upper regime they are comparably lower.
In summary, the identication approach does not substantially inuence the
non-linear reactions to a spending shock. However, changes in a
1
as well as the
overall identication framework have major implications for the GDP response
to revenues. Thus, the exact identication in a structural model is important.
In our view, the PB identication is preferable to a Cholesky ordering, as it is
better suited for distinguishing between the working of automatic stabilisers
and discretionary scal policy.
6.3 Data Sample and Threshold Value
As the observations of 2009 and the end of 2008 are strongly aected by the
nancial crises, we rst re-estimate the model excluding the last 5 periods of the
data sample. The results for the threshold tests do not change signicantly and
are therefore not shown. Furthermore, the shorter data sample yields a similar
threshold estimate of around -0.0015. Figure 18 and 19 show the responses in
the upper and lower growth regime for a shock of 2%. We can nd the main
changes in the lower regime upper right graph, with the responses of revenues
in the rst two quarters being entirely positive. That the main changes occur
in the lower regime is not surprising as the output gap in 2009 was negative
and therefore the last ve observations are covered by the lower regime. The
changes indicate that the eect of spending on revenues was especially strong
27
in the year 2008/2009.
We also re-estimate the model excluding the rst four years of the sample,
starting in 1980 in order to analyse the inuence of the persistently high GDP
growth rates between 1976 and 1980. Since none of the responses shows any
noticeably changes (in neither lower nor upper regime) they are not shown
here.
Furthermore, we conduct the analysis with a higher threshold value to ac-
count for potential inaccuracy in the threshold estimation (although the Tsay
test results are similar for the three dierent lag specications). We employ a
threshold value of zero, which increases the lower regime observations to 74.
The results for the new GIRFs, shown in gure 20 and 21, reveal that only the
upper regime responses change substantially. Government spending as well as
GDP show clear dierences in the reactions to positive and negative revenue
shocks, and the positive as well as the negative scal spending multipliers are
signicantly below zero. Since the lower regime responses do not change signif-
icantly, we can conclude that observations corresponding to a possible middle
regime do not inuence the lower regime, but they lead to a moderation of
the responses in the upper regime.
7 Conclusions
What are the eects of discretionary scal policy shocks? And do they
dier over the dierent phases of the business cycle? In this paper we extend
the existing VAR literature on German scal policy shocks by a non-linear
threshold component, using the output gap as a threshold variable and thereby
dividing the time period from 1976-2009 into a positive and a negative output
gap regime.
In a rst step we estimate a linear benchmark model for which we derive
scal multipliers of around 0.7 (absolute value) for revenue and expenditure
policies, indicating moderate expansionary eects of revenue cuts and expen-
diture increases. These values are supported by the literature, although some
studies derive inverted revenue eects. Those response dierences could result
from diversity in how that data are dened. As such, our revenue series in-
cludes security contributions which are often omitted in the literature (such as
Heppke-Falk et al. 2010). Thus it remains to be seen if we would face similar
problems based on a narrower data denition excluding social security.
As the Tsay (1998) test indicates the necessity of a non-linear model, we
estimate a threshold VAR for a lower (negative output gap) and an upper
(positive output gap) regime. Based on this model we obtain general impulse
response functions that clearly dier between the lower and the upper regime
(and deviate from the linear responses). These deviations have important im-
plications. In periods of a negative output gap, the short-term scal spending
28
multiplier of a positive shock is around unity - indicating a comparatively
high eectivity of economic stimulation via public spending. In contrast, the
short-term spending multiplier for a positive shock found during good times
(positive output gap) is with 0.36 very small, indicating a strong crowding-out
of an expenditure stimulus in booms. The eects of negative spending shocks
dier in both regimes less strongly from the results of the linear model. With
increasing shock size the dierences between positive and negative spending
multipliers and between upper and lower regime increase strongly: The short
term multiplier of a 5% spending increase is 1.27 in the lower but only 0.26 in
the upper regime. Spending reductions of 5% have in both regimes a short-
term multiplier of -0.84. This underlines that the assumption of a linear in-
uence of scal spending on the economy with a multiplier of around 0.7 can
give misleading policy implications. As such, when the output gap is above a
certain threshold, especially expenditure increases could well be less eective
than current linear studies indicate, while our analysis suggests that they are
signicantly more eective in times of a negative output gap. Furthermore
our results show that the dierences between positive and negative shocks in
both regimes increase with the size of the shocks, which further strengthens
the eects described.
With respect to revenue shocks we nd less diverging results than on the
expenditure side. Revenue changes have generally only a limited eect on GDP
with short-term multipliers between 0.48 and 0.62, which dier only slightly
from the multiplier of 0.66 in the linear model. This implies that economic
stimulation in times of negative output gaps works less well via revenue cuts
than via expenditure increases, while the opposite holds for the upper regime.
None of our conclusions changes if we apply a three-quarter moving av-
erage of GDP growth instead of the GDP gap as threshold variable. Fur-
ther robustness checks show that our general implications are not vulnerable
to reasonable changes of the elasticity or the overall structural identication
scheme, the time period analysed or small deviations in the threshold value.
The non-linear threshold analysis shows far more robust behaviour than the
linear analysis, even if the GDP response to revenue shocks is relatively volatile.
Specically, the response dierences between the upper and lower output gap
regime following a spending shock remain statistically signicant. However,
our robustness results re-emphasise the importance of a profoundly deliber-
ated structural identication.
In summary, our analysis suggests that scal steering of the economy via
revenue policies should only (if at all) be pursued in times of a positive output
gap, while discretionary spending measures to boost the economy have a com-
parably larger impact in times of a negative gap and should be concentrated
here. However, our results shall not be interpreted as clear policy advice, they
should rather be understood as indicating gradual dierences in the impact of
scal policy depending on the state of the business cycle.
29
A GIRF Algorithm
Assuming that the non-linear model is known, the GIRF for a given regime
with R observations can be calculated with the following algorithm:
1. Pick a history
r
t1
, with r = 1, ..., R referring to an actual value of
the lagged endogenous variable at a particular date r. Note that R
refers to the values corresponding to the regime the impulse responses
are calculated for. Thus, the same algorithm has to be conducted twice,
for the lower and again for the upper regime.
2. Pick sequences of shocks

t+m
. These are generated by taking bootstrap
samples from the estimated residuals
t
of the TVAR.
3. With the information set
r
t1
, the estimated coecients of the TVAR
and the structural errors

t+m
, simulate the evolution of y over m periods.
The resulting baseline path is given by y
t+m
(
r
t1
|

t+m
).
4. Modify the path of y by adding a shock
0
to the rst residual of the
randomly drawn errors. Again simulate the evolution of y over m periods.
The resulting (shocked) path is given by y
t+m
(
r
t1
|
0
,

t+m
).
5. Repeat steps 2 to 4 B times to get B estimates of the baseline and the
shocked path.
6. Take the average over the dierence of the B estimates of the two paths.
This gives an estimate of the expectation y for a given history
r
t1
.
7. Repeat steps 1 to 6 over all possible histories, that is, the number of
observations R for the regime the GIRF is calculated for.
8. Finally compute the average GIRF for a given regime with R observations
as
y
t+m
(
0
) =
1
R
R

r=1
y
t+m
(
r
t1
|
0
,

t+m
) y
t+m
(
r
t1
|

t+m
)
B
. (24)
With this algorithm, we obtain the GIRFs based on the regime-specic
coecients and contemporaneous coecient matrices resulting from equation
(31).
steps:
30
Table 6: Calculated Elasticities
Elasticity with Average share Weighted
respect to in revenues elasticity
real GDP 1970-2008
Direct taxes
(households and corporations) 1.57 0.27 0.43
Indirect taxes 1 0.27 0.27
Social contributions 0.57 0.42 0.24
Other revenues 0 0.04 0
Elasticity revenues 0 0.94
Unemployment spending -1.4 0.06 -0.08
Elasticity net revenues 1.02
B Exogenous Elasticities
In the literature we nd several methods of calculating exogenous revenue
and expenditure elasticities. For example, Heppke-Falk et al. (2010) derive the
exogenous elasticity based on highly disaggregated time series data, applying
the elasticities calculated by Mohr (2001) and Kremer et al. (2006). We follow
the alternative standard OECD approach (applied by Girouard and Andre
2005, van den Noord 2000, Giorno et al. 1995 and in his scal policy analysis
by Perotti 2004). It comprises a two-step procedure: rst, to calculate the
elasticity of the dierent tax bases and of unemployment with respect to GDP
and then to apply an exogenous elasticity for the reaction of tax revenues to
tax bases and of unemployment spending to unemployment is applied.
The components of the net revenues that are contemporaneously aected
by changes in GDP are direct taxes, indirect taxes, social contributions and
unemployment related spending. Based on the elasticities calculated by the
OECD (see Girouard and Andre 2005) we use a direct tax elasticity of 1.57.
This high elasticity results from the progressive income taxes and the strong
cyclical behavior of corporate prots.
30
Most indirect taxes are levied by pro-
portional rates and have an elasticity of 1. Social contributions increase less
strongly than GDP mainly because they are levied only up to a certain in-
come threshold (which varies depending on the social insurance) and because
the wages as their base react less strongly to GDP than taxable income. The
elasticity of social security contributions in Germany (based on the OECD esti-
mates) is 0.57. If we weigh the individual elasticities pro-rata overall revenues,
the weighed GDP elasticity is on average 0.94.
In contrast to tax revenue, unemployment reacts mirror-inverted to GDP
uctuations and decreases when GDP increases. In the literature we nd a wide
30
The OECD calculates an elasticity of 1.61 for corporate income taxes and 1.53 for
personal income taxes. Because of a methodological break between ESA 1979 and 1995,
there is no consistent separate series for corporate and personal income taxes in our data-
set. Therefore we apply the mean of the two elasticities to all revenue from direct taxes.
31
Table 7: Elasticities in the literature
Elasticity with
respect to Period Data denition
real GDP
Perotti (2004) 0.92 1960-89 net revenues = government revenues
- transfers
- interest
Perotti (2004) 0.91 1960-74 net revenues = government revenues
- transfers
- interest
Perotti (2004) 0.72 1975-89 net revenues = government revenues
- transfers
- interest
Hoppner (2001) 1.04 1970-2000 direct and indirect taxes
Bode et al.(2006) 0.46 1991-2005 taxes and social security contributions
- transfers
Heppke-Falk et al. 0.95 1970-2004 net revenue = government revenues
(2010) - transfers
- interest
Baum/Koester 1.01 1976-2009 net revenue = government and
(2010) social security revenues
- unemployment expenditure
- interest
variation in estimates on the reaction of unemployment to GDP uctuations
in Germany, which range between -5 (Girouard and Andre 2005) and -0.8 (van
den Noord 2000). Based on the German dataset from 1976-2009, we calculate
an elasticity of -1.4. Combining the ratio of unemployment spending over total
revenues (5.7%) with the elasticity of -1.4 and subtracting the resulting value
from the overall revenue elasticity increases the overall net revenue elasticity to
1.02. Thus, a 1% increase of GDP increases net revenues by around 1%. Table
6 summarises the calculation including the average shares of the net revenue
components. To account for the variation of the revenue shares over time, we
use a time-varying elasticity in the structural identication. Instead of the
displayed elasticities calculated based on the average share of the components
over the whole sample, the quarterly elasticities are calculated based on the
share of the components in each respective quarter.
For comparison, table 7 provides elasticities calculated in other German
scal policy studies. The lowest value, at 0.46, is very small (Bode et al.
(2006) using German data covering 1991 to 2005), which results from a lower
eect of GDP on wage growth than the OECD method suggests. Most of the
other papers derive an elasticity that is close to unity (for net revenues). In-
cluding only direct and indirect taxes, the largest elasticity is calculated to be
1.04 (H oppner 2001). Simulation studies for Germany are another point ref-
erence. The values for the eects of automatic stabilisers, derived for instance
32
33
by Toedter and Scharnagl (2004) based on the Bundesbank model, indicate
elasticities which would be closer to 0.5 than to 1. However, these low values
are covered by our robustness tests, which apply an elasticity of 0.5.
C Literature
Afonso, A. and R. M. Sousa (2009): The Macroeconomic Eects of Fiscal
Policy, ECB Working Paper Series, No. 991.
Amisano, G. and C. Giannini, (1997): Topics in Structural VAR Econo-
metrics, Second edition, Springer, Berlin.
Atanasova, C. (2003): Credit Market Imperfections and Business Cy-
cle Dynamics: A Nonlinear Approach, Studies in Nonlinear Dynamics and
Econometrics 7(4), No. 5.
Balke, N. S. (2000): Credit and Economic Activity: Credit Regimes and
Nonlinear Propagation of Shocks, Review of Economics and Statistics 82(2),
pp. 344-349.
Baxter, M. and R. King (1993): Fiscal Policy in General Equilibrium,
American Economic Review 83(3), pp. 315-334.
Blanchard, O. (2003): Macroeconomics, Third Edition, Pearson.
Blanchard, O. and R. Perotti (2002): An Empirical Characterization of
the Dynamic Eects of Changes in Government Spending and Taxes on Out-
put, Quarterly Journal of Economics 117(4), pp. 1329-1368.
Blinder, A. S. (1987): Credit Rationing and Eective Supply Failures,
The Economic Journal 97(386), pp. 327-352.
Bode, O., Gerke, R. and H. Schellhorn (2006): Die Wirkung skalis-
cher Schocks auf das Bruttoinlandsprodukt, Arbeitspapier 01/2006 des
Sachverst andigenrats zur Beurteilung der gesamtwirtschaftlichen Entwicklung.
Breitung, J. B. R. and H. L utkepohl (2004): Structural Vector Autore-
gressive Modelling and Impulse Responses, in H. L utkepohl and M. Kratzig
(eds), Applied Time Series Econometrics, Cambridge University Press, pp.
159-196.
Buiter, Willem H., (1977): Crowding out and the eectiveness of scal
policy, Journal of Public Economics, Elsevier 7(3), pp. 309-328.
Burriel, P., De Castro, F., Garrotte, D. , Gordo, E., Paredes, J. and J.J.
Perez (2010): Fiscal Policy Shocks in the Euro Area and the US: An Empirical
Assessment, Fiscal Studies 31(2), pp. 251-285.
Calza, A. and J. Sousa (2006): Output and Ination Responses to Credit
Shocks: Are There Threshold Eects in the Euro Area?, Studies in Nonlinear
Dynamics and Econometrics 10(2), pp. 1-19.
33
34
Christiano, L. , Aiyagari, S. R. and M. Eichenbaum (1992): The output,
employment, and interest rate eects of government consumption, Journal of
Monetary Economics 30(1), pp. 73-86.
Christiano, L., Eichenbaum M. and S. Rebelo (2009): When Is the Gov-
ernment Spending Multiplier Large?, NBER Working Paper, No. 15394.
Coenen, G., Erceg, C., Freedman, C., Furceri, D., Kumhof, M., Lalonde,
R., Laxton, D., Linde, J., Mourougane, A., Muir, D. , Mursula, S., de Resende,
C., Roberts, J., Roeger, W., Snudden, S. , Trabandt, M. and Jan in

Zt Veld
(2010): Eect of Fiscal Stimulus in Structural Models, IMF Working Paper,
No. 1073.
Coenen, G. and R. Straub (2005): Does Government Spending Crowd in
Private Consumption? Theory and Empirical Evidence for the Euro Area,
International Finance 8(3), pp. 435-470.
Corsetti, G, Kuester, K, Meier, A. and G. J. M uller (2010), Debt Con-
solidation and Fiscal Stabilization of Deep Recessions, American Economic
Review 100(2), pp. 41-45,
Drazen, A. (1991): Can Severe Fiscal Contractions be Expansionary?,
Comment in Blanchard O.J. and S. Fischer (eds.): NBER Macroeconomics
Annual, MIT Press: Cambridge, MA., 1991.
ECB (2007): Government Finance Statistics Guide, Frankfurt/Main.
Edelberg, W., Eichenbaum, M. and J. D. M. Fisher (1999): Understand-
ing the Eects of a Shock to Government Purchases, Review of Economic
Dynamics 2(1), pp. 166-206.
Eichenbaum, M. and J.D.M. Fisher (2005): Fiscal Policy in the Aftermath
of 9/11, Journal of Money, Credit and Banking 37(1), pp. 1-22.
Fat as, A. and L. Mihov (2001): Fiscal Policy and Business Cycles: An
Empirical Investigation, Moneda y Credito 212, pp. 167-210.
Favero C. and F. Giavazzi (2009): How large are the eects of tax
changes?, NBER Working Paper, No. 15303.
Fernandez de Castro, F. and P. Hern andez de Cos (2006): The Economic
Eects of Exogenous Fiscal Shocks in Spain. A SVAR Approach, ECB Work-
ing Paper Series, No. 647.
Giorno, C., Richardson, P. W., Roseveare, D. and P. J. v.d. Noord (1995):
Estimating Potential Output, Output Gaps and Structural Budget Balances,
OECD Economics Department Working Papers.
Gal, J., L opez-Salido, J. D. and J. Valles, (2007): Understanding the
Eects of Government Spending on Consumption, Journal of the European
Economic Association 5(1), pp. 227-270.
34
35
Gallant A. R. , Rossi, P. E. and G. Tauchen (1993): Nonlinear Dynamic
Structures, Econometrica 61(4), pp. 871-907.
Galbraith, J. W. (1996): Credit Rationing and Threshold Eects in the
Relation between Money and Output, Journal of Applied Econometrics 11(4),
pp. 416-429.
Garcia, R. and H. Schaller (2002): Are the Eects of Interest Rate
Changes Asymmetric?, Economic Inquiry 40, pp. 102-119.
Giavazzi, F., Jappelli, T. and M. Pagano (2000): Searching for non-linear
eects of scal policy: Evidence from industrial and developing countries,
European Economic Review, Elsevier 44(7), pp. 1259-1289.
Giavazzi, F. and M. Pagano (1990): Can Severe Fiscal Contractions be
Expansionary? Tales of Two Small European Countries, NBER Working
Paper, No. 3372.
Girouard, N. and C. Andre (2005): Measuring Cyclically-Adjusted Bud-
get Balances for OECD Countries, OECD Economics Department Working
Papers, No. 434.
Gonzalez, M. and J. Gonzalo (1997): Threshold Unit Root Processes,
Unpublished manuscript, Department of Statistics and Econometrics, Univer-
sidad Carlos III de Madrid.
Gonzalo, J. and J. Y. Pitarakis (2002): Estimation and Model Selection
Based Inference in Single and Multiple Threshold Models, Journal of Econo-
metrics 110, pp. 319-352.
Gonzalo, J. and R. Montesinos (2000): Threshold Stochastic Unit Root
Models, Unpublished manuscript, Department of Statistics and Economet-
rics, Universidad Carlos III de Madrid.
Hansen, B. E. (1996): Inference When a Nuisance Parameter Is Not Iden-
tied under the Null Hypothesis, Econometrica 64(2), pp. 413-430.
Hansen, B.E. (1997): Inference in TAR Models, Studies in Nonlinear
Dynamics and Econometrics 2(1), pp. 1-14.
Hansen, B. E. (1999a): Testing for Linearity, Journal of Economic Sur-
veys 13, pp. 551-576.
Hansen, B. E. (1999b): Threshold Eects in Non-Dynamic Panels: Esti-
mation, Testing and Inference, Journal of Econometrics 93(2), pp. 345-368.
Hansen, B. E. (2000): Sample Splitting and Threshold Estimation,
Econometrica 68(3), pp. 575-603.
Hemming R., Kell, M. and S. Mahfouz (2002): The Eectiveness of Fiscal
Policy in Stimulating Economic Activity: A Review of the Literature, IMF
Working Paper, No. 208.
35
36
Heppke-Falk, K. H., Tenhofen, J. and G. B. Wol (2010): The Macroe-
conomic Eects of Exogenous Fiscal Policy Shocks in Germany: A Disaggre-
gated SVAR Analysis, Journal of Economics and Statistics Jahrb ucher f ur
Nationalokonomie und Statistik 230 (3), pp. 328-55.
Hoover, K. D. and S. M. Sherin (1992): Causation, Spending, and Taxes:
Sand in the Sandbox or Tax Collector for the Welfare State?, American Eco-
nomic Review 82(1), pp. 225-248.
Hoppner, F. (2001): A VAR Analysis of the Eects of Fiscal Policy in
Germany, Institute for International Economics Discussion Paper, Bonn.
Hoppner, F. and K. Wesche (2000): Non-linear Eects of Fiscal Policy
in Germany: A Markov-Switching Approach, Bonn Econ Discussion Papers,
9/2000.
Ja askela, J. (2007): More Potent Monetary Policy? Insights from a
Threshold Model, Reserve Bank of Australia, RBA Research Discussion Pa-
pers, 2007-07.
Kn uppel, M. (2008): Can Capacity Constraints Explain Asymmetries of
the Business Cycle?, Bundesbank Discussion Paper Series, 01/2008.
Kollias, C. and S. Paleologou (2006): Fiscal Policy in the European Union:
Tax and Spend, Spend and Tax, Fiscal Synchronisation or Institutional Sepa-
ration?, Journal of Economic Studies 33(2), pp. 108-120.
Koop, G. (1996): Parameter Uncertainty and Impulse Response Analy-
sis, Journal of Econometrics 72(1), pp. 135-49.
Koop, G. and S. M. Potter (1999): Dynamic Asymmetries in US Unem-
ployment, Journal of Business and Economic Statistics 17, pp. 298-312.
Koop, G., Pesaran, M. H. and S. M. Potter (1996): Impulse Response
Analysis in Nonlinear Multivariate Models, Journal of Econometrics 74(1),
pp. 119-147.
Koren, S. and A. Stiassny (1998): Tax and Spend, or Spend and Tax? An
International Study, Journal of Policy Modelling 20, pp. 163-191.
Koske, I. and N. Pain (2008): The Usefulness of Output Gaps for Policy
Analysis, OECD Economics Department Working Papers, No. 621, OECD,
Paris.
Kremer, J., Braz, C. R., Brosens, T., Langenus, G., Momigliano, S., and M.
Spolander (2006): A Disaggregated Framework for the Analysis of Structural
Developments in Public Finances, ECB Working Paper Series, No. 579.
Leibfritz W., Lehne B., Meister, W. and E. Langmantel , (1999): Fi-
nanzpolitik und Konjunktur: die automatischen Stabilisatoren in Deutsch-
land, Ifo-Schnelldienst, pp. 14-22.
36
37
Lo, M. C. and J. Piger (2005): Is the Response of Output to Monetary
Policy Asymmetric? Evidence from a Regime-Switching Coecients Model,
Journal of Money, Credit and Banking 37(5), pp. 865-886.
Lo, M. C. and E. Zivot (2001): Threshold Cointegration and Nonlinear
Adjustment to the Law of One Price, Macroeconomic Dynamics 5, pp. 533-
576.
L utkepohl, H. (2000): Bootstrapping Impulse Responses in VAR Analy-
ses, SFB 373 Discussion Paper Series, 2000-22, Humboldt University, Berlin.
L utkepohl, H. (2005): New Introduction to Multiple Time Series Analysis,
Springer, Berlin Heidelberg New York.
L utkepohl, H. and M. Kratzig (2004): Applied Time Series Econometrics,
Cambridge University Press.
Mohr, M. (2001): Ein disaggregierter Ansatz zur Berechnung konjunk-
turbereinigter Budgetsalden fur Deutschland: Methoden und Ergebnisse,
Bundesbank Discussion Paper Series, 13/2001.
Mountford, A. and H. Uhlig (2009): What Are the Eects of Fiscal Policy
Shocks? Journal of Applied Econometrics, pp. 960-992.
Pagan, A. (1995): Three Econometric Methodologies: An Update, in
Oxley, L. , George, D.A.R., Roberts C.J. and S. Sayer (eds), Surveys in Econo-
metrics, Blackwell, Oxford.
Perotti, R. (2004): Estimating the Eects of Fiscal Policy in OECD Coun-
tries, CEPR Discussion Paper Series, No. 4842.
Perotti, R.(2007): In Search of the Transmission Mechanism of Fiscal
Policy, NBER Working Paper, No. 13143.
Potter, S. M. (2000): Nonlinear Impulse Response Functions, Journal of
Economic Dynamics and Control 24(10), pp. 1425-1446.
Roeger, W. and J. int Veld (2009): Fiscal Policy with Credit Constrained
Households, European Economy, Economic Papers. No. 357.
Romer C. D. and D. Romer (2010): The Macroeconomic Eects of Tax
Changes: Estimates Based on a New Measure of Fiscal Shocks, American
Economic Review, 100(3), pp. 763-801.
Ramey, V and M. Shapiro (1998): Costly Capital Reallocation and the
Eects of Government Spending, Carnegie Rochester Conference on Public
Policy 48, pp. 145-194.
Roos, M. W. M. (2007): Die Makro okonomischen Wirkungen Diskre-
tion arer Fiskalpolitik in Deutschland - Was Wissen Wir Empirisch?, Per-
spektiven der Wirtschaftspolitik 8(4), pp. 293-308.
37
38
Root, T. H. and D. Lien (2003): Impulse Responses in a Threshold Coin-
tegrated System: the Case of Natural Gas Markets, Applied Financial Eco-
nomics 13(1), pp. 23-35.
Spilimbergo, A., Symansky S. A. and M. Schindler (2009): Fiscal Multipli-
ers, IMF Sta Position Note No. 2009/11.
Scharnagl, M. and K.-H. Todter (2004): How Eective are Automatic
Stabilisers? Theory and Empirical Results for Germany and Other OECD
Countries, Deutsche Bundesbank, Discussion Paper Series 1, No. 21.
Tong, H. (1978): On a threshold model, in Chen, C. H.Pattern Recogni-
tion and Signal Processing, Sijtho and Noordho, Amsterdam.
Tong, H. and K. S. Lim (1980): Threshold Autoregression, Limit Cycles
and Cyclical Data, Journal of the Royal Statistical Society, Series B, 42(3),
pp. 245-292.
Tong, H. (1983): Threshold Models in Nonlinear Time Series Analysis,
Springer Lecture Notes in Statistics 21, Springer, New York.
Tsay, R. S. (1998): Testing and Modelling Multivariate Threshold Mod-
els, Journal of the American Statistical Association 93, pp. 1188-1202.
Noord, P. v. d. (2000): The Size and Role of Automatic Stabilisers in
the 1990s and Beyond, OECD Economics Department Working Papers, No.
230, OECD, Paris.
Weise, C. L. (1999): The Asymmetric Eects of Monetary Policy: A
Nonlinear Vector Autoregression Approach, Journal of Money, Credit and
Banking 31(1), pp. 85-108.
Internet: http://www.uni-mannheim.de/edz/pdf/eurostat/06/KS-DT-06-
012-EN.pdf.
38
F
i
g
u
r
e
8
:
L
o
w
e
r
R
e
g
i
m
e
:
a
1
=
0
.
5
,
2
%
G
r
o
w
t
h
S
h
o
c
k
F
i
g
u
r
e
9
:
L
o
w
e
r
R
e
g
i
m
e
:
a
1
=
1
.
5
,
2
%
G
r
o
w
t
h
S
h
o
c
k
39
F
i
g
u
r
e
1
0
:
L
o
w
e
r
R
e
g
i
m
e
:
1
9
7
6
-
2
0
0
8
F
i
g
u
r
e
1
1
:
U
p
p
e
r
R
e
g
i
m
e
:
1
9
7
6
-
2
0
0
8
40
F
i
g
u
r
e
1
2
:
5
%
G
r
o
w
t
h
S
h
o
c
k
-
L
o
w
e
r
R
e
g
i
m
e
F
i
g
u
r
e
1
3
:
5
%
G
r
o
w
t
h
S
h
o
c
k
-
U
p
p
e
r
R
e
g
i
m
e
41
F
i
g
u
r
e
1
4
:
I
R
f
o
r
G
D
P
G
r
o
w
t
h
a
s
T
h
r
e
s
h
o
l
d
:
2
%
G
r
o
w
t
h
S
h
o
c
k
-
L
o
w
e
r
R
e
g
i
m
e
F
i
g
u
r
e
1
5
:
I
R
f
o
r
G
D
P
G
r
o
w
t
h
a
s
T
h
r
e
s
h
o
l
d
:
2
%
G
r
o
w
t
h
S
h
o
c
k
-
U
p
p
e
r
R
e
g
i
m
e
42
F
i
g
u
r
e
1
6
:
L
o
w
e
r
R
e
g
i
m
e
:
T
h
r
e
s
h
o
l
d
o
f
Z
e
r
o
-
2
%
G
r
o
w
t
h
S
h
o
c
k
F
i
g
u
r
e
1
7
:
U
p
p
e
r
R
e
g
i
m
e
:
T
h
r
e
s
h
o
l
d
o
f
Z
e
r
o
-
2
%
G
r
o
w
t
h
S
h
o
c
k
43
F
i
g
u
r
e
1
8
:
L
o
w
e
r
R
e
g
i
m
e
:
C
h
o
l
e
s
k
y
O
r
d
e
r
G
D
P

G
F
i
g
u
r
e
1
9
:
L
o
w
e
r
R
e
g
i
m
e
:
C
h
o
l
e
s
k
y
O
r
d
e
r
R

G
D
P
44
F
i
g
u
r
e
2
0
:
U
p
p
e
r
R
e
g
i
m
e
:
C
h
o
l
e
s
k
y
O
r
d
e
r
G
D
P

G
F
i
g
u
r
e
2
1
:
U
p
p
e
r
R
e
g
i
m
e
:
C
h
o
l
e
s
k
y
O
r
d
e
r
R

G
D
P
45
46
The following Discussion Papers have been published since 2010:
Series 1: Economic Studies
01 2010 Optimal monetary policy in a small open
economy with financial frictions Rossana Merola
02 2010 Price, wage and employment response Bertola, Dabusinskas
to shocks: evidence from the WDN survey Hoeberichts, Izquierdo, Kwapil
Montorns, Radowski
03 2010 Exports versus FDI revisited: C. M. Buch, I. Kesternich
Does finance matter? A. Lipponer, M. Schnitzer
04 2010 Heterogeneity in money holdings Ralph Setzer
across euro area countries: Paul van den Noord
the role of housing Guntram Wolff
05 2010 Loan supply in Germany U. Busch
during the financial crises M. Scharnagl, J. Scheithauer
06 2010 Empirical simultaneous confidence scar Jord, Malte Knppel
regions for path-forecasts Massimiliano Marcellino
07 2010 Monetary policy, housing booms Sandra Eickmeier
and financial (im)balances Boris Hofmann
08 2010 On the nonlinear influence of Stefan Reitz
Reserve Bank of Australia Jan C. Ruelke
interventions on exchange rates Mark P. Taylor
09 2010 Banking and sovereign risk S. Gerlach
in the euro area A. Schulz, G. B. Wolff
10 2010 Trend and cycle features in German
residential investment before and after
reunification Thomas A. Knetsch
47
11 2010 What can EMU countries sovereign
bond spreads tell us about market
perceptions of default probabilities Niko Dtz
during the recent financial crisis? Christoph Fischer
12 2010 User costs of housing when households face Tobias Dmmler
a credit constraint evidence for Germany Stephan Kienle
13 2010 Extraordinary measures in extraordinary times
public measures in support of the financial Stphanie Marie Stolz
sector in the EU and the United States Michael Wedow
14 2010 The discontinuous integration of Western
Europes heterogeneous market for
corporate control from 1995 to 2007 Rainer Frey
15 2010 Bubbles and incentives: Ulf von Kalckreuth
a post-mortem of the Neuer Markt in Germany Leonid Silbermann
16 2010 Rapid demographic change and the allocation
of public education resources: evidence from
East Germany Gerhard Kempkes
17 2010 The determinants of cross-border bank flows
to emerging markets new empirical evidence Sabine Herrmann
on the spread of financial crisis Dubravko Mihaljek
18 2010 Government expenditures and unemployment: Eric Mayer, Stphane Moyen
a DSGE perspective Nikolai Sthler
19 2010 NAIRU estimates for Germany: new evidence
on the inflation-unemployment trade-off Florian Kajuth
20 2010 Macroeconomic factors and Claudia M. Buch
micro-level bank risk Sandra Eickmeier, Esteban Prieto
48
21 2010 How useful is the carry-over effect
for short-term economic forecasting? Karl-Heinz Tdter
22 2010 Deep habits and the macroeconomic effects
of government debt Rym Aloui
23 2010 Price-level targeting C. Gerberding
when there is price-level drift R. Gerke, F. Hammermann
24 2010 The home bias in equities P. Harms
and distribution costs M. Hoffmann, C. Ortseifer
25 2010 Instability and indeterminacy in Michael Krause
a simple search and matching model Thomas Lubik
26 2010 Toward a Taylor rule for fiscal policy M. Kliem, A. Kriwoluzky
27 2010 Forecast uncertainty and the
Bank of England interest rate decisions Guido Schultefrankenfeld
01 2011 Long-run growth expectations M. Hoffmann
and global imbalances M. Krause, T. Laubach
02 2011 Robust monetary policy in a
New Keynesian model with imperfect Rafael Gerke
interest rate pass-through Felix Hammermann
03 2011 The impact of fiscal policy on
economic activity over the business cycle Anja Baum
evidence from a threshold VAR analysis Gerrit B. Koester
49
Series 2: Banking and Financial Studies
01 2010 Deriving the term structure of banking Stefan Eichler
crisis risk with a compound option Alexander Karmann
approach: the case of Kazakhstan Dominik Maltritz
02 2010 Recovery determinants of distressed banks: Thomas Kick
Regulators, market discipline, Michael Koetter
or the environment? Tigran Poghosyan
03 2010 Purchase and redemption decisions of mutual Stephan Jank
fund investors and the role of fund families Michael Wedow
04 2010 What drives portfolio investments of
German banks in emerging capital markets? Christian Wildmann
05 2010 Bank liquidity creation and Berger, Bouwman
risk taking during distress Kick, Schaeck
06 2010 Performance and regulatory effects of
non-compliant loans in German synthetic
mortgage-backed securities transactions Gaby Trinkaus
07 2010 Banks exposure to interest rate risk, their
earnings from term transformation, and
the dynamics of the term structure Christoph Memmel
08 2010 Completeness, interconnectedness and
distribution of interbank exposures
a parameterized analysis of the stability
of financial networks Angelika Sachs
09 2010 Do banks benefit from internationalization? C. M. Buch
Revisiting the market power-risk nexus C. Tahmee Koch, M. Koetter
50
10 2010 Do specialization benefits outweigh Rolf Bve
concentration risks in credit portfolios Klaus Dllmann
of German banks? Andreas Pfingsten
11 2010 Are there disadvantaged clienteles
in mutual funds? Stephan Jank
12 2010 Interbank tiering and money center banks Ben Craig, Goetz von Peter
13 2010 Are banks using hidden reserves Sven Bornemann, Thomas Kick
to beat earnings benchmarks? Christoph Memmel
Evidence from Germany Andreas Pfingsten
14 2010 How correlated are changes in banks net
interest income and in their present value? Christoph Memmel
01 2011 Contingent capital to strengthen the private
safety net for financial institutions:
Cocos to the rescue? George M. von Furstenberg
02 2011 Gauging the impact of a low-interest rate Anke Kablau
environment on German life insurers Michael Wedow
03 2011 Do capital buffers mitigate volatility Frank Heid
of bank lending? A simulation study Ulrich Krger
04 2011 The price impact of lending relationships Ingrid Stein


51
Visiting researcher at the Deutsche Bundesbank


The Deutsche Bundesbank in Frankfurt is looking for a visiting researcher. Among others
under certain conditions visiting researchers have access to a wide range of data in the
Bundesbank. They include micro data on firms and banks not available in the public.
Visitors should prepare a research project during their stay at the Bundesbank. Candidates
must hold a PhD and be engaged in the field of either macroeconomics and monetary
economics, financial markets or international economics. Proposed research projects
should be from these fields. The visiting term will be from 3 to 6 months. Salary is
commensurate with experience.

Applicants are requested to send a CV, copies of recent papers, letters of reference and a
proposal for a research project to:


Deutsche Bundesbank
Personalabteilung
Wilhelm-Epstein-Str. 14

60431 Frankfurt
GERMANY